Joint feature selection and classifier design for radar targets

A joint feature selection and classifier design is proposed in this paper. The approach adopts the feature dimension extension by power transformation as a new kernel function, which can not only make full use of the input samples to form the nonlinear classification boundary, but also realize the nonlinear feature selection. A zero-mean Gaussian prior with Gamma precision is used to promote sparsity in utilization of features in our model. The experiments based on a measured radar data set demonstrate the practicability and effectiveness of the proposed method.